A claim is a place
to start looking.

Snapshot, fork, diff and replay operations are available for supported state. Still open: Coverage and recovery vary by path; external effects and processes are not automatically restored.

Eight modes have composition and coordination services. Still open: Each mode has distinct integration and end-to-end gaps; registration does not establish feature parity.

Twelve runners share comparison, isolated execution, bounded research and persisted resume guards with targeted runtime tests. Still open: The pre-correction R3 campaign ended 12/12 ESCALATE and 0 merge; no corrected-path real-mission campaign is published.

Structured work graphs, independent scheduling, versioned handoffs and program-runner multi-team sub-runs connect to proof-gated completion. Still open: Text-only formation remains lexical; general expertise and generic dispatch outside the qualified program path remain unproven.

Twelve executable variants share persistent deliberation, transactional round receipts, resume and proof-gated finalization. Still open: Models, independent oracles and action adapters remain explicit dependencies. Runtime tests do not establish causal superiority; the dated October 4 campaign reported no verified promotion.

Causal replay, retry fingerprints, persistence rollback and committed-state validation are connected. Still open: Local tests do not establish universal convergence, distributed consensus or general worker correctness.

The provider-backed path resolves, executes, verifies and admits a capability before adding it to persisted routing; the 36-suite Rhizome profile includes an SQLite growth-and-routing E2E. Still open: Hosts must still register real providers and verifiers; composition alone does not start them or prove mission quality.

The shared mission runtime connects capability admission, independent verification, immune gates, quotas and atomic contribution/memory updates; the 55-script suite includes robust host-veto cases. Still open: Physical isolation and measurements remain adapter responsibilities; no general or longitudinal gain is established.

Eleven variants share bounded transactional cycles, authorized adapters, measured consumption and independent verification, with SQLite restart tests. Still open: A global goal requires its own verifier; comparative ecological benefit and real provider quality remain unqualified.

The autonomous regional runtime persists reviewed migration, recolonization, local evidence and recovery; 14 topology suites and a minimal external adapter qualification are recorded. Still open: Global qualification and the quality of host-provided evaluators, receivers and solvers remain open.

All nineteen declared organizations now have deterministic runOrganizationStep coverage and executable browser counterparts on this site. Still open: A guidance policy is not autonomous dispatch, and no general improvement in mission outcomes follows from algorithm coverage alone.

Registered plugins can execute a validated graph on the explicit opt-in path. Still open: Execution receipts are unverified traces; graph completion does not prove a mission outcome or apply a transition.

The inventory covers 19 WorkerKind values, 25 methods and 24 benchmark cases; all 19 types have C2/C3 evidence on one bounded task, with independent oracles and persisted mission receipts. Still open: This does not establish correctness on arbitrary missions, complete qualification of every method, or broad parity with other agent systems.

JS stdio and backend dispatch enforce leases and direct calls. Still open: Rust parity and individual tool nominal/refusal coverage vary by tool.

Web observation and foveation handlers exist. Still open: A complete observation–action–verification mission loop is not published.

The Node Omega gateway executes six guarded operations with persisted session visibility, MMU resolution and verification receipts. Still open: General free-form mission compilation, complete Rust production bindings and empirical PGO qualification remain open.

A kernel and targeted routing are integrated. Still open: General mission integration is pending; relation constraints cannot grant new authority.

Persisted deliveries, an inbox and ACK states, cognitive escalation and leased handler retries are connected. Still open: Receiver registration is not wired at server startup; delivery is at least once and external actions require idempotency.

Completion, feedback, dormancy and regeneration have connected paths, including a two-process SQLite WAL authority test. Still open: A multi-instance deployment has not been qualified.

Transactional experiment waves connect verified coverage, Trinity hypotheses and Biome niches under explicit contracts. Still open: Host adapters must ensure real isolation and verification; empirical advantage is not qualified.

Bounded counterfactual search connects authorized interventions, verified history and snapshot restoration. Still open: Host restoration and general mission outcomes remain domain-specific and need comparative evidence.

Declarative procedures support isolated compression replay, transactional consolidation, scoped recall and transitive invalidation. Still open: Known-case preservation does not establish general agent-memory safety or coverage of future failures.

Resident ticks dispatch read-only probes and persist observed coverage, failures and missed windows. Still open: Creating a schedule does not start a daemon; field detection and coverage remain unqualified.

Preregistered evaluation units, inherited risk grants, provenance checks and signed receipts guard protected topology promotions. Still open: Valid individual statistical tests and field evidence remain required; these gates do not replace deterministic safety checks.

Native numerical procedures, POET snapshots, cultural reuse and persisted phenotype run with SQLite reopen and six executed ablation arms. Still open: Open-ended creativity, arbitrary user tasks and all 25 conceptual mechanisms are not demonstrated; search costs are not equalized across arms.

Versioned roots, immutable bases, private overlays, signed inventory, explicit materialization and candidate ingestion are connected to opted-in supervised missions. Still open: No head is published automatically; general promotion still requires a path-specific verifier and real runtime evidence.

All 420 registry entries map to an explicit bounded audit over eleven shared primitives; missing or violated observations create verification work instead of success. Still open: These audits do not validate philosophical theories, external truth, subjective consciousness or general mission benefit.

A deterministic, read-only chain detects an internal signal, formulates and triages a falsifiable question, queues it through Ontogenesis and persists a bounded unverified proposal. Still open: The concept document specifies independent classification, but the verifier service and test are absent from the pinned published tree. Learning and the reserved comparative campaign are also not implemented; ETT is not a test of phenomenal consciousness.